| From: | Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> |
|---|---|
| To: | Casey Duncan <casey(at)pandora(dot)com> |
| Cc: | Andrew Sullivan <ajs(at)crankycanuck(dot)ca>, pgsql-hackers(at)postgresql(dot)org |
| Subject: | Re: Hints (Was: Index Tuning Features) |
| Date: | 2006-10-13 01:09:18 |
| Message-ID: | 21792.1160701758@sss.pgh.pa.us |
| Views: | Raw Message | Whole Thread | Download mbox | Resend email |
| Thread: | |
| Lists: | pgsql-hackers |
Casey Duncan <casey(at)pandora(dot)com> writes:
> Yes, but it may be much more efficient for the human to tell the
> computer than for the computer to introspect things. Take, for
> example, ndisinct as data grows large.
Yeah, an override estimate for a column's ndistinct seems a perfect
example of the sort of statistical hint that I'd be in favor of having.
We have also talked about solving the multi-column statistics problem
(which, at its core, is "which combinations of columns are worth
accumulating stats for?" --- you can't possibly store stats for every
combination!) by having what would amount to hints from the DBA saying
"keep stats for these combinations".
regards, tom lane
| From | Date | Subject | |
|---|---|---|---|
| Next Message | Andrew Dunstan | 2006-10-13 02:01:50 | Re: Hints (Was: Index Tuning Features) |
| Previous Message | Andrew Dunstan | 2006-10-13 01:01:06 | Re: ./configure argument checking |